search

rinongal / StyleGAN-nada

http://stylegan-nada.github.io/
MIT License
1.16k stars 146 forks source link

Question with quantitative evaluation on image diversity #46

Closed jone1222 closed 2 years ago

jone1222 commented 2 years ago

Hello Rinongal,

As far as I know, you mentioned that you clustered around generated images using K-medoids since yours does not use a training set. How did you set the number of clusters K? (the few-shot adaptation methods choose K to be the number of training samples)

Can I ask you to provide details on evaluating image diversity quantitatively? If you can provide the source code for quantitative evaluation, I would be grateful. Thanks.

rinongal commented 2 years ago

Hi jone,

We used 10 clusters to align with the level of supervision that the most recent methods at the time (Ojha et al.) were aimed at.

Regarding the code, here's my current uncommented file: eval_diversity.txt It's based on Ojha's code.

Be sure to convert the extension to .py (github doesn't allow uploading py files as part of the response...).

You can flip between the k_med and data-centered versions by just commenting out the lines at the bottom (and commenting out the kmeds version). Let me know if you have any issues with it and I'll try to help / clean up the script.